Synthetic Wind Turbine Dataset

Dataset for the paper "Synthetic Imagery Aided Geographic Domain Adaptation for Rare Energy Infrastructure Detection in Remotely Sensed Imagery".
The dataset consist of 4 folders which contains imagery and labeling files for 4 geographic domains: Northeast (NE), Northwest (NW), Eastern Midwest (EM), and Southwest(SW). Each domain folder then has 3 sub-folders: "Real_training", "Real_validation", and "Synthetic_training" which represent real imagery training set, real imagery validation set, and synthetic imagery training set respectively as mentioned in the paper.
Within each subfolder, the ".jpg" or ".png" files are images while the ".txt" files are labeling files. An image matches its corresponding labeling file by filename (excluding file extensions).
E.g., "EM/Real_training/naip_2513_IA_WND_i0j0.jpg" is a image file in the real imagery training set from EM domain. The "EM/Real_training/naip_2513_IA_WND_i0j0.txt" file under the same subfolder is the corresponding labeling file for the image. If an image file has no matching labeling (".txt") file in the same subfolder, that image file has no wind turbines within and no labeling is needed.
To reproduce the results from the paper "Synthetic Imagery Aided Geographic Domain Adaptation for Rare Energy Infrastructure Detection in Remotely Sensed Imagery", the code base in https://github.com/Duke-BC-DL-for-Energy-Infrastructure/Experiment-Setup is recommended (instructions available in the linked page)